Phonotactic rules in speech perception

Elliott Moreton

This paper presents experimental evidence 1. that speakers of English
know the phonotactics of their language, 2. that they use this
knowledge when processing speech input, and 3. that phonotactic
knowledge consists of categorical, rule-like prohibitions, rather than
emerging from statistical properties of the lexicon.

In every language, some sequence of sounds are illegal.  English, for
instance, bans lax high vowels word-finally -- [bI] cannot be an
English word.  Linguists traditionally attribute this to
language-particular phonotactic rules.  More recently, some
psychologists have suggested instead that phonotactics is an emergent
statistical property of the lexicon, caused by the frequency of some
sequences and the rarity of others in the English vocabulary.  The
issue is tested here by comparing the effects of absolute phonotactic
rules versus nonphonotactic lexical frequency differences on phonetic
category boundaries.

Stimuli are disyllabic English pseudowords ending in a stressed
syllable whose vowel is ambiguous.  One continuum is [gri] (very
common in that position) to [grI] (illegal); another is [kri] (legal
but very rare) to [krI] (illegal).  Controls, with both endpoints
legal, are [gridZ]-[grIdZ] and [kridZ]-[krIdZ].  The [I] endpoint's
illegality was found to move the [I]-[i] boundary towards [I] compared
with controls.  The rule theory predicts equal shifts for the [gr] and
[kr] continua; the statistical theory says it will be larger for the
frequent [gr].  Equal-sized shifts were found.  

[Abstract in:  Journal of the Acoustical Society of America, 102 (5,
Part 2): 3091-3092 (1997).]